Speaker
Description
Large datasets are becoming increasingly common in astronomy, with LSST Rubin and Euclid, poised to release an immense wealth of data. To process this data, astronomers have developed tools such as outlier detection algorithms to probe for novel detections and information. In this presentation, we utilise an unsupervised machine learning model, an Isolation Forest (IF), to pinpoint outlying E+A galaxies from GAMA DR4. From our IF model, we report a range of rare astronomical detections and phenomena that have not been highlighted by other surveys, demonstrating the robustness of the IF in finding outliers. We find evidence for E+A galaxies that – although selected in a “normal” manner using H𝛅 – are still star-forming, with extremely strong absorption lines, but little to no [OII] or [OIII] forbidden lines. Of the 48 most unusual E+A galaxies, we report that 25% are possible star formers and more than half are found in large cluster groups. We discuss these findings in the context of what it means to use these selection criteria and how it may affect previous findings in this area. We further discuss the strength of the IF model in finding outliers, its ability in discovering data reduction errors and its capability of being applied to a wide selection of surveys and samples.
